Insurance Bayesian Network: A Discrete Medium-Scale Probabilistic Model

A Bayesian network with 27 nodes and 52 arcs, modeling insurance-related variables. The network contains 1008 parameters and was created by J. Binder, D. Koller, S. Russell, and K. Kanazawa, with a reference publication from 1997.

Use Cases

• Learning and inference on Bayesian network structures based on the 27-node graph.
• Benchmarking probabilistic reasoning algorithms using the defined network parameters.
• Studying the properties of medium-scale discrete networks, such as average Markov blanket size and degree.
• Teaching concepts of directed acyclic graphs and conditional probability tables in an insurance context.

Strengths

• Well-defined network structure with 27 nodes and 52 arcs, providing a concrete model for analysis.
• Detailed topological metrics provided, including average Markov blanket size (5.19) and maximum in-degree (3).
• Clear academic provenance with authors and a citable 1997 Machine Learning publication.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for certain statistical tasks.
• The dataset's specific insurance variables and their states are not described in the provided metadata.

Provenance

Source

bnlearn Bayesian Network Repository (discrete-medium section).

Collection Method

Likely constructed as a synthetic or canonical example for probabilistic network research.

Time Range

Publication reference is from 1997; dataset creation date is unknown.

Freshness

Last update date is unknown; freshness unverified.

Geography

Spatial coverage is unknown.